!pip install numpy pandas matplotlib seaborn
!pip install scikit-learn
import matplotlib.pyplot as plt
import seaborn as sns
import pandas as pd
import numpy as np
from scipy.stats import zscore
from sklearn import linear_model
from sklearn.model_selection import train_test_split
from sklearn.metrics import mean_squared_error

Requirement already satisfied: numpy in c:\users\austin\anaconda3\lib\site-packages (1.26.4)
Requirement already satisfied: pandas in c:\users\austin\anaconda3\lib\site-packages (2.2.2)
Requirement already satisfied: matplotlib in c:\users\austin\anaconda3\lib\site-packages (3.8.4)
Requirement already satisfied: seaborn in c:\users\austin\anaconda3\lib\site-packages (0.13.2)
Requirement already satisfied: python-dateutil>=2.8.2 in c:\users\austin\anaconda3\lib\site-packages (from pandas) (2.9.0.post0)
Requirement already satisfied: pytz>=2020.1 in c:\users\austin\anaconda3\lib\site-packages (from pandas) (2024.1)
Requirement already satisfied: tzdata>=2022.7 in c:\users\austin\anaconda3\lib\site-packages (from pandas) (2023.3)
Requirement already satisfied: contourpy>=1.0.1 in c:\users\austin\anaconda3\lib\site-packages (from matplotlib) (1.2.0)
Requirement already satisfied: cycler>=0.10 in c:\users\austin\anaconda3\lib\site-packages (from matplotlib) (0.11.0)
Requirement already satisfied: fonttools>=4.22.0 in c:\users\austin\anaconda3\lib\site-packages (from matplotlib) (4.51.0)
Requirement already satisfied: kiwisolver>=1.3.1 in c:\users\austin\anaconda3\lib\site-packages (from matplotlib) (1.4.4)
Requirement already satisfied: packaging>=20.0 in c:\users\austin\anaconda3\lib\site-packages (from matplotlib) (23.2)
Requirement already satisfied: pillow>=8 in c:\users\austin\anaconda3\lib\site-packages (from matplotlib) (10.3.0)
Requirement already satisfied: pyparsing>=2.3.1 in c:\users\austin\anaconda3\lib\site-packages (from matplotlib) (3.0.9)
Requirement already satisfied: six>=1.5 in c:\users\austin\anaconda3\lib\site-packages (from python-dateutil>=2.8.2->pandas) (1.16.0)
Requirement already satisfied: scikit-learn in c:\users\austin\anaconda3\lib\site-packages (1.4.2)
Requirement already satisfied: numpy>=1.19.5 in c:\users\austin\anaconda3\lib\site-packages (from scikit-learn) (1.26.4)
Requirement already satisfied: scipy>=1.6.0 in c:\users\austin\anaconda3\lib\site-packages (from scikit-learn) (1.13.1)
Requirement already satisfied: joblib>=1.2.0 in c:\users\austin\anaconda3\lib\site-packages (from scikit-learn) (1.4.2)
Requirement already satisfied: threadpoolctl>=2.0.0 in c:\users\austin\anaconda3\lib\site-packages (from scikit-learn) (2.2.0)

# Introduction
# In this project, we'll be looking at wind turbines, and creating regression models based off of various features. The most important question I want 
# to answer with this project is: "Can we create a model that can accurately predict the power generated by wind turbines?", I would also like to 
# explore and see how the various features play a role in the predicting power of our models. I think that this will be influential and more developed 
# forms of the models that I intend to create here could actually be useful to companies that deal with wind turbines specifically, such as helping 
# in determining which features to look out for when building wind turbines, as well as companies that deal with other potential sources of
# energy where the information is applicable. 

# Our dataset comes from Kaggle, and can be found here: https://www.kaggle.com/datasets/sudhanvahg/wind-turbine-power-generation-forecasting
# This dataset includes data on the hourly meteorological data along with power output measurements from wind turbines collected between Jan 2013 
# and December 2016. The data is split into two portions, a train set and a test set. We'll only be working with the training set, as the test set
# does not include a column for the "power" feature, which I intend to be our target in our models. The included variables are temperature at 2 meters,
# relative humidity at 2 meters, dew point at 2 meters, wind speed at different altitudes (10 meters, 100 meters), and wind direction for these
# altitudes as well. There are 140160 entries in our training dataset. There are no null values.

# What is Regression?
# Regression is a way to predict unknown values based on values we already know based on the trends and relationships within the data. Linear regression
# fits lines to our existing data, picks the best one, and then uses that line to predict future values that we don't already know. It's like making an
# educated guess!

# How we'll be evaluating our models:
# One of the ways to measure how good of a model we have is the R-squared value of a model. The R-squared value is a ratio of the residual sum of 
# squared divided by the total sum of squares. Here's what the scores mean in layman's terms: R-Squared values closer to 0 mean the values predicted
# by the model are as good as randomly guessing, while values closer to 1 are like we already knew the values prior to predicting them.
# Another score we'll be using to measure how good our model's predictions are is RMSE, or Root Mean Square Error, which tells us how far off our
# predicted values are from the actual values. The closer the RMSE score is to 0, the more accurate our model's predictions are.

wt_train_df_raw = pd.read_csv("train.csv")
wt_train_df_raw.describe()

wt_train_df_raw.isna().sum()

Unnamed: 0    0
Time          0
Location      0
Temp_2m       0
RelHum_2m     0
DP_2m         0
WS_10m        0
WS_100m       0
WD_10m        0
WD_100m       0
WG_10m        0
Power         0
dtype: int64

# Preprocessing
# We'll start by dropping the "Unnamed: 0", Location, and "Time" columns. The "Unnamed: 0" column is just an index column, so we won't need that.
# The "Time" column could potentially hold a lot of valuable information if we were to break it down by season, and the time of day. Unfortunately for
# my own sake, I'll be choosing to not explore this option for this project - I don't think that the juice is worth the squeeze or that the difference 
# between times of day and seasons will produce different enough results from looking at everything alltogether. Additionally, we'll
# remove the outliers from our data. This can be accomplished by removing entries that are over 3 standard deviations from the mean (average) of our
# data set. The reason we're choosing that number is because approximately 99.7% of our data should lie within 3 standard deviations of the mean.
# Removing these entries will help prevent our data from skewing the results of our models, and since our dataset has over 100,000 entries, we 
# should still have plenty of information to work with.

wt_train_df = wt_train_df_raw.drop('Unnamed: 0', axis=1)
wt_train_df = wt_train_df.drop('Time', axis = 1)
wt_train_df.describe()

# Data Understanding/Visualization
sns.pairplot(wt_train_df)

<seaborn.axisgrid.PairGrid at 0x130e3007dd0>

# Based on the above pairplot, it does look like we have a few outliers here and there, especially in our windspeed graphs, so let's get started
# in dealing with our outliers. We can take the z-score from our data, which is a measurement that describes a value's relationship relative to 
# our data's mean (average). It is measured in standard deviations from the mean, so this makes it perfect for identifying if a value is an outlier.
# We'll then remove values that are more than 3 standard deviations from the mean.

# To explain what we're doing below, we take the absolute value of each z-score, and we're removing those entries which have values that have z-scores
# that are more than 3 standard deviations from the mean of our data in either direction (below or above).
# First we need to identify the columns we're going to take the z-score of. We won't be taking the z-score of the location, because it doesn't really
# apply since it's more of a categorical value than anything else.
columns_to_score = ['Temp_2m','RelHum_2m','DP_2m','WS_10m','WS_100m','WD_10m','WD_100m','WG_10m','Power']
# Then we'll take the absolute value of the zscores for the entries, because it doesn't matter if it is above or below as long as it's more than 3 std 
# dev away from the mean.
z_scores = np.abs(wt_train_df[columns_to_score].apply(zscore))
# We'll then filter the z-scores of our data to exclude any values that are greater than 3 standard deviations from the mean.
filtered_rows = (z_scores < 3).all(axis=1)
# We'll now update our data set with what we're keeping.
wt_train_df_flt = wt_train_df[filtered_rows]
wt_train_df_flt.describe()

# We've now cut our dataframe down from 140160 entries to 138519 entries - meaning we've removed about 1641 entries that were outliers
# from our data. Awesome!

# Model 1
# We'll start by separating our target, the "Power" column, from the rest of our features.
X = wt_train_df_flt.drop(labels = 'Power', axis = 1)
y = wt_train_df_flt['Power']

# Next we split our data into test and train sets.
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size = .2, random_state = 42)
print("X_train", X_train.shape)
print("X_test", X_test.shape)
print("y_train", y_train.shape)
print("y_test", y_test.shape)

X_train (110815, 9)
X_test (27704, 9)
y_train (110815,)
y_test (27704,)

# We're going to use a Linear Regression model this time. We're going to do this because it seems like it would be perfect for what we're doing, which is
# predicting unknown target values based on known values.
wt_linear_model = linear_model.LinearRegression()
wt_linear_model.fit(X_train, y_train)

LinearRegression()

LinearRegression()

# Now to predict and see how well our model performs!
y_pred = wt_linear_model.predict(X_train)
# This gets us the R-Squared score of our model, which we previously talked about.
print(wt_linear_model.score(X_train, y_train))
# And this generates the RMSE score of our model.
print(np.sqrt(mean_squared_error(y_train, y_pred)))

0.4961092927436167
0.178789351836086

# Currently, our model has an R-Squared value of about .49, which isn't great. Typically, the closer a value is to 1 the better, and our score right
# now isn't looking so hot. In our next experiment, we'll attempt to remedy this by removing some of the columns that aren't as correlated to our power
# generated, which we'll use a correlation matrix to gauge.
# On the flip side, our RMSE score is actually pretty solid! It's surprisingly close to 0, so our model does pretty well at predicting close to the actual
# power values!

# Experiment 2
# For the second experiment I'd like to try removing a few of our features that aren't really correlated with our target feature.
# Now let's take a peek at the relationships between our values, to do this, we'll be using a correlation matrix.
sns.heatmap(wt_train_df_flt.corr(), annot = True)
plt.show()

# Our features that have the highest correlation to the power generated are the Wind Speed at 10m, Wind Speed at 100m, and Wind Gusts at 10m.
# Because of this, we'll try another Linear Regression model using only wind speeds at 10 and 100 meters, and wind gusts at 10m. This means we'll be 
# dropping the columns for Location, Temperature, Relative Humidity, Dew Point, and wind direction at 10 and 100m.

# We'll start by dropping the previously mentioned columns, then we'll separate our target column once more.
wt_train_df_flt2 = wt_train_df_flt.drop('Temp_2m', axis = 1)
wt_train_df_flt2 = wt_train_df_flt2.drop('Location', axis = 1)
wt_train_df_flt2 = wt_train_df_flt2.drop('RelHum_2m', axis = 1)
wt_train_df_flt2 = wt_train_df_flt2.drop('DP_2m', axis = 1)
wt_train_df_flt2 = wt_train_df_flt2.drop('WD_10m', axis = 1)
wt_train_df_flt2 = wt_train_df_flt2.drop('WD_100m', axis = 1 )
X2 = wt_train_df_flt2.drop(labels = 'Power', axis = 1)
y2 = wt_train_df_flt2['Power']

# We'll also split up our data into test and train sets once more.
X_train2, X_test2, y_train2, y_test2 = train_test_split(X2, y2, test_size = .2, random_state = 42)
print("X_train2", X_train2.shape)
print("X_test2", X_test2.shape)
print("y_train2", y_train2.shape)
print("y_test2", y_test2.shape)

X_train2 (110815, 3)
X_test2 (27704, 3)
y_train2 (110815,)
y_test2 (27704,)

# Time to create our model and fit data to it!
wt_linear_model2 = linear_model.LinearRegression()
wt_linear_model2.fit(X_train2, y_train2)
# Let's evaluate it now.
y_pred2 = wt_linear_model2.predict(X_train2)
print(wt_linear_model2.score(X_train2, y_train2))
print(np.sqrt(mean_squared_error(y_train2, y_pred2)))

0.4647408276373267
0.18427038128302245

# Sadly, our model now produces ever-so-slightly worse predictions with our R-Squared score dropping by about .03, and our RMSE score increasing by about
# .01. I didn't expect this because I figured that our score would improve with less bloat from lesser correlated features in the mix. This suggests
# there may be another factor at play, such as the location of our turbines. Let's look at that next!

# Experiment 3
# I want to separate our data in order to look at the power generated by the turbines at each separate location.

# We'll backtrack to our filtered data set (the one where we've already dropped outliers, time, and unnamed: 0 columns) and split it into the 
# 4 different turbine locations.
wt_1 = wt_train_df_flt[wt_train_df_flt['Location']== 1]
wt_2 = wt_train_df_flt[wt_train_df_flt['Location']== 2]
wt_3 = wt_train_df_flt[wt_train_df_flt['Location']== 3]
wt_4 = wt_train_df_flt[wt_train_df_flt['Location']== 4]

# Wind Turbine Location 1
wt_1.describe()

# We've got bout 35,000 entries that are from Location 1, so this should be enough to build a decent model!
# We'll do the same process for all of the following models, where we split the target from the data, create train and test sets at an 80/20 ratio, 
# create the model, and print our results. We'll evaluate each individually, then take a look at the bigger picture!
X = wt_1.drop(labels = 'Power', axis = 1)
y = wt_1['Power']
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size = .2, random_state = 42)
print("X_train", X_train.shape)
print("X_test", X_test.shape)
print("y_train", y_train.shape)
print("y_test", y_test.shape)
wt_1_lm = linear_model.LinearRegression()
wt_1_lm.fit(X_train, y_train)
y_pred = wt_1_lm.predict(X_train)
print("R-Squared: ", wt_1_lm.score(X_train, y_train))
print("RMSE: ", np.sqrt(mean_squared_error(y_train, y_pred)))

X_train (27857, 9)
X_test (6965, 9)
y_train (27857,)
y_test (6965,)
R-Squared:  0.6183843354245468
RMSE:  0.17551267418407876

# This model performs significantly better than our initial model, this may mean that some locations are dragging down our results, but for now we'll
# hold that thought. Our RMSE is about on par with our initial model as well, so this model is performs really well!

# Wind Turbine Location 2
wt_2.describe()

# We've got bout 34,000 entries that are from Location 2, so this should be sufficient to build a model!
X = wt_2.drop(labels = 'Power', axis = 1)
y = wt_2['Power']
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size = .2, random_state = 42)
print("X_train", X_train.shape)
print("X_test", X_test.shape)
print("y_train", y_train.shape)
print("y_test", y_test.shape)
wt_2_lm = linear_model.LinearRegression()
wt_2_lm.fit(X_train, y_train)
y_pred = wt_2_lm.predict(X_train)
print("R-Squared: ", wt_2_lm.score(X_train, y_train))
print("RMSE: ", np.sqrt(mean_squared_error(y_train, y_pred)))

X_train (27392, 9)
X_test (6848, 9)
y_train (27392,)
y_test (6848,)
R-Squared:  0.44295334128844366
RMSE:  0.15583336498184827

# This model has an R-Squared value of .44, which is slightly worse than that of our initial model. This one could be a culprit in our initial model's
# poorer performance. However, it does have a slightly lower RMSE value - which means it's predictions were closer to the actual values.

# Wind Turbine Location 3
wt_3.describe()

# We've got bout 35,000 entries that are from Location 3, so this should be sufficient to build a model as well.
X = wt_3.drop(labels = 'Power', axis = 1)
y = wt_3['Power']
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size = .2, random_state = 42)
print("X_train", X_train.shape)
print("X_test", X_test.shape)
print("y_train", y_train.shape)
print("y_test", y_test.shape)
wt_3_lm = linear_model.LinearRegression()
wt_3_lm.fit(X_train, y_train)
y_pred = wt_3_lm.predict(X_train)
print("R-Squared: ", wt_3_lm.score(X_train, y_train))
print("RMSE: ", np.sqrt(mean_squared_error(y_train, y_pred)))

X_train (27724, 9)
X_test (6931, 9)
y_train (27724,)
y_test (6931,)
R-Squared:  0.389298113035603
RMSE:  0.19589308398025804

# This model's R-Squared score is even lower than that of turbine 2's model. Yikes. The RMSE score of this model was also worse than the initial model's.
# Double yikes.

# Wind Turbine Location 4
wt_4.describe()

# We've got bout 35,000 entries that are from Location 4, so this should be sufficient to build a model as well.
X = wt_4.drop(labels = 'Power', axis = 1)
y = wt_4['Power']
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size = .2, random_state = 42)
print("X_train", X_train.shape)
print("X_test", X_test.shape)
print("y_train", y_train.shape)
print("y_test", y_test.shape)
wt_4_lm = linear_model.LinearRegression()
wt_4_lm.fit(X_train, y_train)
y_pred = wt_4_lm.predict(X_train)
print("R-Squared: ", wt_4_lm.score(X_train, y_train))
print("RMSE: ", np.sqrt(mean_squared_error(y_train, y_pred)))

X_train (27841, 9)
X_test (6961, 9)
y_train (27841,)
y_test (6961,)
R-Squared:  0.6968848783083041
RMSE:  0.12446564544282326

# The model for location 4 performed well! Its R-Squared value is good, and it actually has the best RMSE score out of all of our models. Awesome, now
# let's look at what all of this means.

# Once we separated the models by location, we saw that locations 2 and 3 produce much lower scores in comparison to locations 1 and 4. Location seems
# to play a huge role in the correlation of the features versus the power generated by wind turbines. This makes perfect sense because if a turbine is 
# placed in a terrible location, it would produce power less consistently and less efficiently than a turbine placed in an average or exceptional 
# location. It also could suggest some issues with our model, such as issues with the fit of the line, issues identifying trends/patterns, or it 
# could suggest that our initial model was overfitting to some locations, and underfitting to others. There could also be missing features that we
# just don't know about too, such as ages of these wind turbines, anything about where the location actually are located, like on top of a hill or 
# in a valley, or something else completely.

# Impact:
# This project has a lot of potential for impact within the realm of alternative/renewable energy sources. More in depth models than my own could play
# extremely important roles in deciding the locations of future Wind Turbines. Typically in the United States, wind turbines are placed in wide open 
# areas, like at the coast or in the plains of the western regions of our country. Models such as these could help cherry pick the best of these 
# locations, and increase efficiency greatly! These sort of models could help prevent waste of resources in building new turbines as well. If a model
# can provide relatively accurate information as to what we can expect from a turbine in a new location, it would help those in charge determine whether
# or not a new turbine would actually be worth it to build a wind turbine there. A potential issue with models like mine could be if we assume that these
# turbines are operating at a constant. As anyone who has been outside knows, the wind is not a constant thing. Some days there may be no wind at all, 
# and other days the wind may be strong enough to knock someone over. This would have to be taken into consideration as well with any actual model.
# Another thing that would need to be considered is the environmental impact of more turbines. Any piece of machinery is guaranteed to fail at some
# point or other. It is not a question of if, but when. There are countless instances of fires, or other forms of mechanical failure that have already
# occurred with wind turbines, with one as recently as January, 2024 in Colorado. These sorts of failures can lead to disasters such as wildfires,
# especially in more arid places, and loss of life drastically affecting ecosystems. These are all things that should be considered, and with proper 
# maintenance and care can be mitigated at the very least.

# Conclusion:
# To answer our initial question, yes, we can indeed create a model for wind turbines that accurately predicts power generated based off of other
# features. To get a decently accurate model, we had to split it up by the specific locations and bring the information down to a finer scale, instead of
# a more general prediction, but I feel satisfied with what we came up with. The biggest factor in improving the model I'd have to say was splitting up
# the model by location, which increased our R-Squared scores by upwards of 20% (.20), as well as helped identify what was dragging down the initial
# model. I also saw that just removing features from a model just because they aren't as correlated to the target value doesn't necessarily mean that 
# the model will perform better. We saw this when in the second experiment the R-Squared score actually went down instead of up when we removed values
# that had lower values in our correlation matrix.

# I've learned a lot from working on this project about Linear Regression, R-Squared values, and most definitely the importance of features and the
# roles that they play in our models. Just because something has a lesser correlation, doesn't mean that it isn't still valuable information to have, and
# when broken down into more parts, may even paint a more clear picture of what is actually going on. I've also learned that I really enjoy tinkering
# and tweaking these models, adjusting what features are included and aren't, and thinking through the logic of why something should or shouldn't be 
# included in a model. Doing projects like this, and exploring logic and choices such as those detailed throughout the project can prove to be very
# valuable, and help me become better overall when working with data, be it in the logic of building a model, or in the actual coding portion. 
# Specific to this project, I really enjoyed the subject matter of the data. However, I did feel like the dataset was lacking in the information on
# the write up when I reflect on it. Some of the features were unclear, especially location. The initial data set on kaggle didn't seem to include any
# information on the specific location of these wind turbines, they could've been spread out across the US, the world, or across a single field, and we
# wouldn't know. Features like these could potentially play a huge role in the meaning and outcomes of the projects. I think my biggest take away in 
# this aspect is that I seriously need to take more time and go through these data sets with a fine tooth comb when I'm selecting them. 

# Overall, I'm mostly satisfied with the work I've done in this project, but I wish I could've spent more time tinkering with the models, and had more
# features to work with. In the future, I'll try to mitigate this as best as possible on my end (where my other classes and available datasets allow
# haha).

# References
# Kaggle Data Set: https://www.kaggle.com/datasets/sudhanvahg/wind-turbine-power-generation-forecasting

	Unnamed: 0	Location	Temp_2m	RelHum_2m	DP_2m	WS_10m	WS_100m	WD_10m	WD_100m	WG_10m	Power
count	140160.000000	140160.000000	140160.000000	140160.000000	140160.000000	140160.000000	140160.000000	140160.000000	140160.000000	140160.000000	140160.000000
mean	17519.500000	2.500000	45.912162	70.420512	35.867228	4.352948	6.924278	200.684417	200.649010	8.027673	0.312437
std	10115.212797	1.118038	21.930554	17.000203	20.979720	2.027149	3.056636	100.079917	101.105919	3.626641	0.253774
min	0.000000	1.000000	-31.420400	8.664205	-36.627405	0.165389	0.007799	0.051683	-0.942685	0.436515	-0.000004
25%	8759.750000	1.750000	29.579600	57.664205	21.172595	2.815389	4.747799	128.051683	127.057315	5.236515	0.099696
50%	17519.500000	2.500000	45.879600	72.664205	35.672595	4.055389	6.717799	211.051683	211.057315	7.636515	0.246896
75%	26279.250000	3.250000	64.579600	84.664205	53.872595	5.575389	8.847799	287.051683	289.057315	10.236515	0.486396
max	35039.000000	4.000000	94.479600	99.664205	78.272595	18.695389	24.597799	359.051683	359.057315	28.936515	0.988796

	Location	Temp_2m	RelHum_2m	DP_2m	WS_10m	WS_100m	WD_10m	WD_100m	WG_10m	Power
count	140160.000000	140160.000000	140160.000000	140160.000000	140160.000000	140160.000000	140160.000000	140160.000000	140160.000000	140160.000000
mean	2.500000	45.912162	70.420512	35.867228	4.352948	6.924278	200.684417	200.649010	8.027673	0.312437
std	1.118038	21.930554	17.000203	20.979720	2.027149	3.056636	100.079917	101.105919	3.626641	0.253774
min	1.000000	-31.420400	8.664205	-36.627405	0.165389	0.007799	0.051683	-0.942685	0.436515	-0.000004
25%	1.750000	29.579600	57.664205	21.172595	2.815389	4.747799	128.051683	127.057315	5.236515	0.099696
50%	2.500000	45.879600	72.664205	35.672595	4.055389	6.717799	211.051683	211.057315	7.636515	0.246896
75%	3.250000	64.579600	84.664205	53.872595	5.575389	8.847799	287.051683	289.057315	10.236515	0.486396
max	4.000000	94.479600	99.664205	78.272595	18.695389	24.597799	359.051683	359.057315	28.936515	0.988796

	Location	Temp_2m	RelHum_2m	DP_2m	WS_10m	WS_100m	WD_10m	WD_100m	WG_10m	Power
count	138519.000000	138519.000000	138519.000000	138519.000000	138519.000000	138519.000000	138519.000000	138519.000000	138519.000000	138519.000000
mean	2.501281	46.117744	70.478512	36.092496	4.282069	6.828647	200.037186	199.993453	7.910036	0.309281
std	1.120388	21.803271	16.955317	20.827574	1.903810	2.911459	100.063676	101.088483	3.446334	0.251757
min	1.000000	-19.820400	19.664205	-26.927405	0.165389	0.007799	0.051683	-0.942685	0.436515	-0.000004
25%	1.000000	29.679600	57.664205	21.272595	2.805389	4.717799	127.051683	126.057315	5.236515	0.098696
50%	3.000000	46.279600	72.664205	35.872595	4.015389	6.677799	210.051683	210.057315	7.536515	0.244196
75%	4.000000	64.679600	84.664205	54.072595	5.515389	8.767799	287.051683	288.057315	10.236515	0.480096
max	4.000000	94.479600	99.664205	78.272595	10.425389	16.067799	359.051683	359.057315	18.836515	0.985896

	Location	Temp_2m	RelHum_2m	DP_2m	WS_10m	WS_100m	WD_10m	WD_100m	WG_10m	Power
count	34822.0	34822.000000	34822.000000	34822.000000	34822.000000	34822.000000	34822.000000	34822.000000	34822.000000	34822.000000
mean	1.0	47.247293	71.991728	37.796224	3.750078	6.287366	201.480664	201.216697	7.689691	0.402823
std	0.0	19.467290	16.748881	18.822979	1.602365	2.620822	96.974314	98.511891	3.451752	0.284251
min	1.0	-14.620400	19.664205	-17.527405	0.165389	0.107799	0.051683	-0.942685	0.436515	-0.000004
25%	1.0	31.779600	59.664205	23.472595	2.585389	4.407799	127.051683	126.057315	4.936515	0.150796
50%	1.0	46.279600	73.664205	37.372595	3.475389	6.107799	223.051683	224.057315	7.236515	0.347346
75%	1.0	63.679600	85.664205	53.872595	4.765389	8.007799	276.051683	277.057315	9.936515	0.651196
max	1.0	93.879600	99.664205	75.872595	9.825389	15.527799	359.051683	359.057315	18.836515	0.985696

	Location	Temp_2m	RelHum_2m	DP_2m	WS_10m	WS_100m	WD_10m	WD_100m	WG_10m	Power
count	34240.0	34240.000000	34240.000000	34240.000000	34240.000000	34240.000000	34240.000000	34240.000000	34240.000000	34240.000000
mean	2.0	42.146756	70.315899	32.113539	4.754643	7.432421	206.394206	204.404044	8.299371	0.260148
std	0.0	23.445184	16.701965	21.102631	2.027169	3.138597	103.897288	105.683065	3.401441	0.208648
min	2.0	-19.820400	19.664205	-26.927405	0.165389	0.147799	0.051683	-0.942685	0.436515	-0.000004
25%	2.0	25.079600	58.664205	18.072595	3.165389	5.107799	130.051683	125.057315	5.636515	0.086696
50%	2.0	42.179600	72.664205	32.072595	4.575389	7.357799	224.051683	217.057315	8.036515	0.207096
75%	2.0	62.279600	83.664205	50.372595	6.155389	9.647799	298.051683	301.057315	10.636515	0.393996
max	2.0	92.379600	99.664205	76.172595	10.425389	16.047799	359.051683	359.057315	18.836515	0.985896

	Location	Temp_2m	RelHum_2m	DP_2m	WS_10m	WS_100m	WD_10m	WD_100m	WG_10m	Power
count	34655.0	34655.000000	34655.000000	34655.000000	34655.000000	34655.000000	34655.000000	34655.000000	34655.000000	34655.000000
mean	3.0	48.715722	68.532939	37.825370	4.463788	6.996482	194.178446	194.238962	8.031183	0.321752
std	0.0	22.318090	17.533032	21.926877	1.970482	2.932109	102.204960	102.527774	3.516491	0.251020
min	3.0	-19.720400	19.664205	-26.527405	0.165389	0.007799	0.051683	-0.942685	0.436515	-0.000004
25%	3.0	31.279600	54.664205	22.172595	2.915389	4.887799	125.051683	125.057315	5.336515	0.104596
50%	3.0	49.579600	69.664205	37.772595	4.215389	6.907799	191.051683	193.057315	7.636515	0.268796
75%	3.0	67.879600	83.664205	57.472595	5.755389	8.957799	289.051683	289.057315	10.336515	0.513196
max	3.0	94.479600	99.664205	78.272595	10.425389	16.017799	359.051683	359.057315	18.836515	0.977196

	Location	Temp_2m	RelHum_2m	DP_2m	WS_10m	WS_100m	WD_10m	WD_100m	WG_10m	Power
count	34802.0	34802.000000	34802.000000	34802.000000	34802.000000	34802.000000	34802.000000	34802.000000	34802.000000	34802.000000
mean	4.0	46.307406	71.061768	36.576940	4.168473	6.609089	198.172509	200.160326	7.626823	0.251605
std	0.0	21.274989	16.631046	20.820647	1.843725	2.808181	96.657156	97.232681	3.371100	0.226369
min	4.0	-19.620400	19.664205	-26.927405	0.165389	0.007799	0.051683	0.057315	0.536515	-0.000004
25%	4.0	30.079600	58.664205	21.372595	2.725389	4.607799	126.051683	127.057315	5.036515	0.068696
50%	4.0	46.279600	72.664205	36.672595	3.935389	6.507799	206.051683	210.057315	7.236515	0.188696
75%	4.0	64.679600	84.664205	54.647595	5.365389	8.447799	283.051683	285.057315	9.836515	0.372096
max	4.0	92.179600	99.664205	77.472595	10.425389	16.067799	359.051683	359.057315	18.836515	0.978396